CN104205805A - Image processing device and method, and image processing program - Google Patents
Image processing device and method, and image processing program Download PDFInfo
- Publication number
- CN104205805A CN104205805A CN201380017115.5A CN201380017115A CN104205805A CN 104205805 A CN104205805 A CN 104205805A CN 201380017115 A CN201380017115 A CN 201380017115A CN 104205805 A CN104205805 A CN 104205805A
- Authority
- CN
- China
- Prior art keywords
- value
- pixel
- pass filter
- coefficient
- low
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012545 processing Methods 0.000 title claims abstract description 79
- 238000000034 method Methods 0.000 title claims description 18
- 238000001914 filtration Methods 0.000 claims abstract description 39
- 238000012937 correction Methods 0.000 claims description 56
- 238000012360 testing method Methods 0.000 claims description 22
- 230000033228 biological regulation Effects 0.000 claims description 7
- 238000003672 processing method Methods 0.000 claims description 7
- 230000002093 peripheral effect Effects 0.000 abstract 1
- 230000015654 memory Effects 0.000 description 14
- 238000011946 reduction process Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 9
- 238000001514 detection method Methods 0.000 description 7
- 238000003384 imaging method Methods 0.000 description 5
- 239000003086 colorant Substances 0.000 description 4
- 230000001915 proofreading effect Effects 0.000 description 3
- GZPBVLUEICLBOA-UHFFFAOYSA-N 4-(dimethylamino)-3,5-dimethylphenol Chemical compound CN(C)C1=C(C)C=C(O)C=C1C GZPBVLUEICLBOA-UHFFFAOYSA-N 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000926 separation method Methods 0.000 description 2
- 101100453811 Caenorhabditis elegans kgb-1 gene Proteins 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
Classifications
-
- G06T5/70—
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/60—Noise processing, e.g. detecting, correcting, reducing or removing noise
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration by the use of local operators
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/80—Camera processing pipelines; Components thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20182—Noise reduction or smoothing in the temporal domain; Spatio-temporal filtering
Abstract
An NR coefficient (Kf) is calculated on the basis of a representative value of a target pixel and the low-pass filter value nearest to the representative value of the target pixel, said low-pass filter value being selected from among a plurality of different-direction one-dimensional low pass filter values corresponding to an input image, and the pixel value of the target pixel and a peripheral low-pass filter (25) are weighted by the NR coefficient (Kf) and added (36). The representative value of the target pixel might be, for example, a median value obtained by filtering a current frame and a past frame in the time axis direction. By this means, noise is reduced in the image processing device without lowering the resolution.
Description
Technical field
The present invention relates to remove image processing apparatus, method and the image processing program of sneaking into the noise in input picture in digital camera etc.
Background technology
Thering is the image-input device such as digital camera of the imaging apparatuss such as CCD and showing from the image output devices such as the display of the image of this image-input device input, in image input system or image delivery system, in view data, sneak into noise.Generally in device, remove this noise by image processing.The existing method of removing sneaked into noise in this image-input device or image output device is described.
As the easiest conventional method of removing noise, there is following technology: form the processing window being formed by multiple pixels centered by the pixel as image handling object (hereinafter referred to as concerned pixel), be weighted addition by the pixel value of the neighboring pixel in processing window and the pixel value of concerned pixel, apply low-pass filtering.But, in the situation that using the method, for the significant edge existing in the image beyond noise, also apply without exception low-pass filtering, therefore, there is the problem of the decrease resolution of image.
And, as additive method, following technology is proposed: for processing window, by applying Pu Ruiweite (Prewitt) filtering or Sobel (Sobel) filtering carrys out Edge detected, the significant edge in image in the case of being judged as concerned pixel, weaken low pass filtered intensity of wave, or applying low-pass filtering along being present in the direction at edge of periphery, thus, in the decrease resolution that suppresses image, remove noise (for example, with reference to patent documentation 1).
Prior art document
Patent documentation
Patent documentation 1: TOHKEMY 2002-185795 communique
Summary of the invention
The problem that invention will solve
In the method for recording at above-mentioned patent documentation 1, sometimes noise erroneous judgement continuous on specific direction be decided to be to edge or the pattern erroneous judgement existing in subject is decided to be to noise, causing the reduction of resolution thereby exist, and can not fully carry out the problem of noise reduction.
The present invention has put in order to address the above problem just, and its object is, can carry out noise reduction and can not reduce resolution.
For solving the means of problem
Image processing apparatus of the present invention is characterised in that, described image processing apparatus is using the dynamic image being made up of the time series of multiframe as input picture, described image processing apparatus has: multiple one dimension low pass filters, they calculate the low-pass filter values of the pixel that uses described concerned pixel and upwards arrange at mutual different one-dimensional square respectively with respect to described concerned pixel in the processing window being made up of multiple pixels centered by the concerned pixel of paying close attention to frame; Periphery low pass filter, it uses the low-pass filter values of described concerned pixel and surrounding pixel thereof for described processing window output, as periphery low-pass filter values; Typical value calculating part, it,, according to the pixel value of the pixel of position corresponding with described concerned pixel in 1 beyond the pixel value of the described concerned pixel of described concern frame and described concern frame above frame, obtains the typical value relevant with described concerned pixel; Noise level test section, its calculate one-dimensional square described in each to low-pass filter values and the absolute value of the difference of described typical value, in the absolute value of the described difference that output calculates, meet the minimization of the sum of absolute value of rated condition, as the noise level of concerned pixel; NR coefficient generating unit, it receives described noise level, generates the noise reduction coefficient corresponding with this noise level according to predefined input-output characteristic; And NR handling part, it uses pixel value and the described periphery low-pass filter values of described noise reduction coefficient to described concerned pixel to be weighted addition, in described NR handling part, described noise reduction coefficient is larger, larger for the weight of described periphery low-pass filter values.
Invention effect
According to the present invention, do not distinguish pixel value variation and noise that original edge causes, can adjust the mixing ratio of the pixel value of periphery low-pass filter values and concerned pixel, therefore, can carry out noise reduction and can not reduce resolution, can obtain the image of high image quality.
Brief description of the drawings
Fig. 1 is the block diagram that the image processing apparatus of embodiments of the present invention 1 is shown.
Fig. 2 is the key diagram that the method for the filtering processing of the 1LPF of execution mode 1 is shown.
Fig. 3 is the key diagram that the method for the filtering processing of the 2LPF of execution mode 1 is shown.
Fig. 4 is the key diagram that the method for the filtering processing of the 4LPF of execution mode 1 is shown.
Fig. 5 is the key diagram that the method for the filtering processing of the 4LPF of execution mode 1 is shown.
Fig. 6 is the key diagram that the method for the filtering processing of the two-dimentional LPF of execution mode 1 is shown.
Fig. 7 is the line chart that an example of the input-output characteristic of the NR coefficient generating unit 34 of execution mode 1 is shown.
(a) of Fig. 8 and (b) be the key diagram that the picture signal before and after the noise reduction process of execution mode 1 is shown.
Fig. 9 is the block diagram that the image processing apparatus of embodiments of the present invention 2 is shown.
Figure 10 is the line chart that an example of the input-output characteristic of the correction coefficient generating unit of execution mode 2 is shown.
Figure 11 is the block diagram that the image processing apparatus of embodiments of the present invention 3 is shown.
Figure 12 is the line chart that an example of the characteristic of the 1st interpolation coefficient of the noise level for the 2nd little direction of execution mode 3 and the noise level of minimum direction is shown.
Figure 13 is the figure that an example of the characteristic of the 2nd interpolation coefficient of the noise level for minimum direction of execution mode 3 is shown.
Embodiment
Execution mode 1
Fig. 1 illustrates the functional structure of the image processing apparatus of embodiments of the present invention 1.
The dynamic image that the time series by multiframe that illustrated image processing apparatus is for example supplied with the image input systems such as illustrated imaging apparatus never or image delivery system forms is as input picture.Image is rectangular arrangement in (line direction) and vertical direction (longitudinal direction) pixel by the horizontal direction forms, input image data by arrange according to raster order the pixel that represents composing images pixel value data and the data that obtain form.
Below, suppose following situation: input picture is the image of the single-plate color digital camera input of the pixel from possessing the color filter with Bayer arrangement, input successively the picture signal of the shooting color of 4 colors of R, Gr, Gb, B according to putting in order of pixel from imaging apparatus.
Illustrated image processing apparatus has line storage 10, the 1st frame memory 11 and the 2nd frame memory the 12, the 1st one dimension low pass filter 21～4th one dimension low pass filter 24 (below low pass filter being called to LPF), two-dimensional low-pass filter (LPF) 25, median filter 30, noise level test section 32, NR coefficient generating unit 34, NR handling part 36.
Line storage 10 receives input image data P (x, y) and keeps the view data of regulation line number.Here, regulation line number N is for example 17.
Pixel in center row during the N that selects successively line storage 10 to keep is capable is as concerned pixel, centered by the concerned pixel of selecting by rectangular area capable longitudinal N, laterally M pixel as processing window, the determined pixel in this processing window is output to respectively the 1st one dimension LPF21～4th one dimension LPF24 and two-dimentional LPF25.
In the example of following explanation, M and N are both 17.
The input image data of the concerned pixel of reading from line storage 10 roughly has the delay of (N/2) row with respect to the input image data that is input to line storage 10, but content is identical, therefore utilizes identical symbol P (x, y) to represent.
The 1st frame memory 11 and the 2nd frame memory 12 have respectively the capacity of accumulating 1 frame image data, with acting on the delay portion that makes the data of the concerned pixel in data that line storage 10 keeps postpone respectively for 1 image duration to export.Consequently, the view data from the 1st frame memory 11 is exported 1 frame of present frame, the view data from the 2nd frame memory 12 is exported 2 frames of present frame.
One dimension LPF21～24 are for the processing window being made up of level 17 pixels, vertical 17 pixels centered by the concerned pixel by input image data, calculate respectively the LPF value La (x of different one-dimensional squares to (0 degree, 45 degree, 90 degree, 135 degree), y), Lb (x, y), Lc (x, y), Ld (x, y).
, the low-pass filter values of the pixel that uses concerned pixel and upwards arrange at mutual different one-dimensional square respectively with respect to concerned pixel, in the processing window being made up of multiple pixels centered by the concerned pixel of paying close attention to frame, is calculated in multiple LPF21～24.
Fig. 2 illustrates the pixel using in the filtering of 0 degree direction is processed for processing window.The pixel that is positioned at center is concerned pixel P (x, y).In 1LPF21, use the pixel (applying the pixel of oblique line) in the upper arrangement of 0 degree direction (horizontal direction) that comprises concerned pixel P (x, y) to carry out filtering.
The pixel that forms input picture has the color filter that Bayer is arranged, and in the situation that carrying out filtering processing for each concerned pixel, only uses the pixel of the same color in concerned pixel pixel (pixel in window) around.Below 2LPF22～the 4LPF24 and the two-dimentional LPF25 of explanation are also same.
In 2LPF22, as shown in Figure 3, the pixel (applying the pixel of oblique line) in the upper arrangement of 45 degree directions (being rotated counterclockwise the direction after 45 degree with respect to horizontal direction) that use comprises concerned pixel P (x, y) is carried out filtering.
In 3LPF23, as shown in Figure 4, use the pixel (applying the pixel of oblique line) in the upper arrangement of 90 degree directions (vertical direction) that comprises concerned pixel P (x, y) to carry out filtering.
In 4LPF24, as shown in Figure 5, the pixel (applying the pixel of oblique line) in the upper arrangement of 135 degree directions (being rotated counterclockwise the direction after 135 degree with respect to horizontal direction) that use comprises concerned pixel P (x, y) is carried out filtering.
In each one dimension LPF21～24, the one-dimensional square that for example comprises concerned pixel by calculating to pixel value simple average of pixel (pixel using in filtering), calculate the average pixel value La (x of all directions, y), Lb (x, y), Lc (x, y), Ld (x, y), respectively as one-dimensional square to LPF value export.
Two dimension LPF25 uses the pixel in the processing window centered by concerned pixel to calculate the LPF value of two-dimensional directional.In execution mode 1, two-dimentional LPF25 is periphery low pass filter.
Fig. 6 utilizes oblique line that the pixel using in the LPF computing for processing window is shown.In the example shown in Fig. 6, as the LPF value of two-dimensional directional, use concerned pixel in the processing window being formed by 17 pixel × 17 pixels and the pixel with concerned pixel same color, calculate their whole average value P f (x, y) according to formula (1).
In formula (1), (x, y) pays close attention to the coordinate of pixel.
About the coordinate of each pixel, the upper left corner of establishing image is (0,0), in right, often advance 1 pixel separation and make x increase by 1, below upwards often advance 1 pixel separation and make y increase by 1.
Median filter 30 calculates pixel value (signal level) P (x, y) of the concerned pixel of the present frame the view data of exporting from line storage 10
t, concerned pixel 1 frame exported from the 1st frame memory 11 pixel value P (x, y)
t-1, concerned pixel 2 frames exported from the 2nd frame memory 12 pixel value P (x, y)
t-2median (median) Pm (x, y), the typical value that sets it as concerned pixel is exported.
Form typical value calculating part by median filter 30.
, typical value calculating part 30 is according to the pixel value P (x, y) of the concerned pixel of concern frame
tpixel value with the pixel of position corresponding with concerned pixel in 1 above frame paying close attention to beyond frame, carries out the filtering of time-axis direction, obtains thus the typical value relevant with concerned pixel.
And typical value is that the pixel by using concerned pixel and the position corresponding with it is carried out the value that the filtering of time-axis direction obtains.And here, typical value is median Pm (x, y).
In addition, in the situation that not needing to distinguish frame, omit the small tenon " t " for determining present frame.
Noise level test section 32 calculate each one-dimensional square of being calculated by one dimension LPF21～24 respectively to LPF value La (x, y)～Ld (x, y) with the typical value Pm (x of the concerned pixel exported from median filter 30, absolute value ADa (the x of difference y), y), ADb (x, y), ADc (x, y), ADd (x, y), minimization of the sum of absolute value ADmin (the x of the difference that output calculates, y) as the noise level NL (x, y) of concerned pixel.
; noise level test section 32 calculate each one-dimensional squares to low-pass filter values La (x; y)～Ld (x; y) with typical value Pm (x; absolute value ADa (x, y), the ADb (x, y) of difference y), ADc (x; y), ADd (x; absolute value ADa (x, y), the ADb (x of the difference that y), output calculates; y), ADc (x; y), in ADd (x, y), meet the minimization of the sum of absolute value Dmin (x, y) of rated condition; as the noise level NL (x, y) of concerned pixel.
Absolute value ADa (x, y), the ADb (x, y) of above-mentioned difference, ADc (x, y), ADd (x, y) are respectively by with following formula (2a)～(2d) performance.
ADa(x，y)＝|La(x，y)-Pm(x，y)|…(2a)
ADb(x，y)＝|Lb(x，y)-Pm(x，y)|…(2b)
ADc(x，y)＝|Lc(x，y)-Pm(x，y)|…(2c)
ADd(x，y)＝|Ld(x，y)-Pm(x，y)|…(2d)
In noise level test section 32, export the minimum value ADmin (x, y) in absolute value ADa (x, the y)～ADd (x, y) of above-mentioned difference, as the noise level NL (x, y) of concerned pixel.
NR coefficient generating unit 34 receives the noise level NL (x, y) being detected by noise level test section 32, generates noise reduction coefficient (NR coefficient) Kf (x, y) according to predefined input-output characteristic.
That is, NR coefficient generating unit 34 receives noise level NL (x, y), generates the noise reduction coefficient Kf (x, y) corresponding with noise level NL (x, y) according to predefined input-output characteristic.
Fig. 7 illustrates an example of the input-output characteristic of NR coefficient generating unit 34.In illustrated example, the pixel value of input image data is got 0～4095 value, correspondingly, supposes that the noise level detecting also gets 0～4095 value.
Show following characteristic for the noise level detecting: the scope of 65 stage definitions NR coefficients with 0～64, the noise level ADmin (x, y) detecting is larger, the degree of noise reduction is stronger, noise level ADmin (x, y) is less, and the degree of noise reduction is more weak.
In the example shown in Fig. 7, at noise level NL (x, y) while being less than DMIN, NR COEFFICIENT K f (x, y) be fixed on DMINY, at noise level NL (x, y) be in the scope of DMIN～DMAX time, NR COEFFICIENT K f (x, y) is increased to DMAXY along with the increase of noise level NL (x, y) from DMINY, at noise level NL (x, y), when exceeding in the scope of the 2nd setting DMAX, NR COEFFICIENT K f (x, y) is fixed on DMAXY.
DMINX, the DMAXX of regulation input-output characteristic, DMINY, DMAXY are according to the shooting condition in image-input device and predefined.
In NR handling part 36, use the NR COEFFICIENT K f (x being calculated according to each pixel by NR coefficient generating unit 34, y), as shown in following formula (3), to the pixel value P (x of concerned pixel, y) be weighted addition with the LPF value Pf (x, y) of the two-dimensional directional relevant with this concerned pixel.
Pb(x，y)
＝(Kf(x，y)×Pf(x，y)+(1-Kf(x，y))×P(x，y))/64
…(3)
, NR handling part 36 uses pixel value P (x, y) and the periphery low-pass filter values Pf (x, y) of noise reduction coefficient Kf (x, y) to concerned pixel to be weighted addition.
Shown in (3), utilize the LPF value Pf (x of NR COEFFICIENT K f (x, y) to two-dimensional directional, y) be weighted, utilize (64-Kf (x, y)) to be weighted the pixel value P (x, y) of concerned pixel.
The pixel value Pb (x, y) of the value Pb (x, y) that carries out the computing of formula (3) and obtain after as noise reduction process and exporting from NR handling part 36.
By carrying out this processing, noise level NL (the x obtaining for each pixel, y)=ADmin (x, y) larger, the NR COEFFICIENT K f (x, y) relevant with each pixel is larger, two-dimentional LPF value Pf (x, y) mixing ratio is larger, and noise reduction (NR) effect is stronger.The larger expression of minimum value ADmin (x, y) of difference absolute value is not the directive edge of variation tool of pixel value, but the variation of pixel value does not have directive noise.
That is, in NR handling part 36, noise reduction coefficient Kf (x, y) is larger, larger for the weight of periphery low-pass filter values Pf (x, y).
About above-mentioned a succession of processing, these 4 colors of taking in color of R, Gr, Gb, B of processing window have only been described, still, carry out too above-mentioned a succession of processing for other each colors.
Carry out respectively same processing in order to take color for 4, can distinguish (according to each shooting color) arranges Fig. 1 separately LPF21～25, median filter 30, noise level test section 32, NR coefficient generating unit 34 and NR handling part 36, also can utilize same circuits to process in order.
By carrying out as mentioned above noise reduction process, can realize picture entirety noise level homogenizing and can not damage resolution, therefore, can obtain the noise reduction of high image quality.
(a) of Fig. 8 and the example of the picture signal before and after above-mentioned noise reduction process (b) is shown.For easy and one-dimensional signal is shown.
(a) of Fig. 8 illustrates the picture signal of the image processing apparatus that is input to Fig. 1.In figure, the variation of the pixel value that originally had of signal indication subject that dotted line is recorded, still, the impact of the noise of sneaking in the analog circuit such as imaging apparatus, amplifying circuit due to the input side of image processing apparatus and as the signal input of solid line.
(b) of Fig. 8 illustrates that the signal of (a) to Fig. 8 according to the input-output characteristic of Fig. 7 has carried out the picture signal after noise reduction process.
Can confirm according to Fig. 8 (a) with (b), noise signal is inhibited, and on the other hand, the edge originally comprising in image is maintained.This be because, each one-dimensional square that calculates 0 degree～135 degree in noise level test section 32 to LPF value and the difference absolute value of the typical value of concerned pixel, by the noise level NL (x using its minimum value as concerned pixel, y), accurately detection noise level, carries out noise reduction process with the intensity corresponding with the noise level detecting.
As mentioned above, according to this execution mode 1, for the processing window being formed by multiple pixels centered by concerned pixel, the different one-dimensional square of difference that calculating comprises concerned pixel to multiple LPF values, and, calculate the LPF value of the two-dimensional directional that comprises concerned pixel for above-mentioned processing window, calculate the typical value between multiframe for concerned pixel, calculate each one-dimensional square to LPF value and the difference absolute value of typical value, minimum value in the difference absolute value that detection computations goes out is as the noise level NL (x of concerned pixel, y), generate and noise level NL (x according to predefined input-output characteristic, y) corresponding NR COEFFICIENT K f (x, y), the NR COEFFICIENT K f (x generating by use, y) the pixel value P (x to concerned pixel, and the LPF value Pf (x of two-dimensional directional y), y) be weighted addition, carry out noise reduction.
Like this, owing to being configured to accurately detection noise level, therefore, can carrying out noise reduction and can not reduce resolution, can obtain the image of high image quality.And then, obtain such as median of typical value by reference time direction of principal axis, noise remove performance and Edge preservation performance improve, and, because the pixel count of reference on time-axis direction is Min., therefore, can be realized by small-scale circuit.
And, about the noise level NL (x of concerned pixel, y) detection, the noise that is configured to unborn edge in differentiate between images not and added by camera system carries out level inhibition, therefore, the structure of only noise being carried out to noise reduction with using edge in the differentiate between images such as Sobel filter with noise is compared, and can export the image that granular sense is even and noise level is low.
And, according to execution mode 1, the median of each pixel value of the concerned pixel before concerned pixel, 1 frame of the present frame being calculated by median filter 30 and the concerned pixel before 2 frames is only for the detection of noise level, and is not used in the weighting summation in NR handling part 36.By such formation, in image after noise reduction process, only use the pixel of present frame, in the case of between photographing before present frame and 1 frame exist motion subject, by only using the pixel of present frame, also can carry out noise reduction process and exist motion subject in can not produce shake.
In addition, in above-mentioned example, for whole shooting colors of the image that uses single-plate color transducer to photograph, generate NR coefficient according to unified input-output characteristic, but, the invention is not restricted to this, also can be configured to according to each shooting color (each color component) of input picture and use different input-output characteristics.Thus, according to the characteristic of the image input system of each color component or image delivery system, can carry out best noise reduction according to the characteristic of input picture.
; input picture P (x; y) comprise multiple color components; NR coefficient generating unit 34 is according to input picture P (x; y) each color component; utilize the input-output characteristic corresponding with the characteristic of input picture P (x, y) to generate the noise reduction coefficient Kf (x, y) of each color component.
And, input picture P (x, y) comprise multiple color components, correction coefficient Kf (x, y) generate according to each color component of input picture P (x, y), utilize according to input picture P (x, y) characteristic and different input-output characteristic generates correction coefficient Kf (x, y).Here, correction coefficient is noise reduction coefficient Kf (x, y).
And, in above-mentioned example, in one dimension low pass filter 21～24, use is positioned at whole pixels of the same color of arranging in prescribed direction of processing window and calculates LPF value, in two-dimensional low-pass filter 25, use whole pixels of the same color in processing window to calculate LPF value, but, the invention is not restricted to this, also can be configured to, only select to have the pixel value of the pixel value closer with the median being calculated by median filter 30 (that is, being positioned at specified level poor), for calculating LPF value.Thus, the in the situation that of comprising feature and the visibly different region of concerned pixel in processing window, the computing of the feature of zones of different can be got rid of, more high-resolution noise reduction can be realized.
That is, in processing window, only select with the difference of typical value be positioned at regulation level difference pixel value and obtain each low-pass filter values, thereby obtain one dimension low pass filter 21～24 and periphery low pass filter 25.
And, in above-mentioned example, by noise level test section 32 detection noise level NL (x, y) time, estimate each one-dimensional square to LPF value and the difference absolute value of the typical value (median) of concerned pixel in minimum value as noise level NL (x, y), but, the invention is not restricted to this, also can be configured to, use minimum value and maximum in the LPF value that the 1st one dimension LPF21～4th one dimension LPF24 calculates, in the case of the filtering direction corresponding with minimum value be with the orthogonal direction of filtering direction corresponding to maximum, be genuine edge direction by the filtering direction determining corresponding with minimum value, be noise level by this minimum estimation.
In this situation, for example, in the time that filtering direction corresponding to the filtering direction discord corresponding with minimum value and maximum is orthogonal, replace minimum value and judge filtering direction corresponding to the value little with the 2nd whether and orthogonal with filtering direction corresponding to maximum, if orthogonal, the 2nd little value is processed as noise level.If filtering direction corresponding to the value little with the 2nd also got along well orthogonal with filtering direction corresponding to maximum, investigate the 3rd little value.Generally speaking, investigate successively the little value of N (N is more than 2 integer), the difference absolute value that meets initial and corresponding with the maximum orthogonal such condition of filtering direction is processed as noise level.Even if also do not meet and when the filtering direction orthogonal such condition corresponding with maximum, return to original rule to N in investigation, minimum value is processed as noise level.
Above processing can be described as following processing: from multiple one-dimensional squares to LPF value extract the LPF value that meets the difference absolute value minimum in the LPF value of rated condition and between typical value, and be used as noise level.
Like this, by carrying out from the processing of multiple edge direction detection noise level NL (x, y), select the possibility of the noise level absolute value of the opposite way round and reduce due to the impact of random noise, can calculate noise level more accurately.
And, in above-mentioned example, calculate the NR coefficient being calculated by NR coefficient generating unit 34 with 65 stages, still, the invention is not restricted to this, the number of stages of NR coefficient can be according to factors such as the bit widths of input picture and arbitrary decision.
And, in above-mentioned example, pixel value by concerned pixel and pixel around thereof simple on average calculate each one-dimensional square to LPF value and the LPF value of two-dimensional directional, but, the invention is not restricted to this, also can be configured to, give the weight corresponding with distance to concerned pixel for each surrounding pixel of concerned pixel and calculate on average.Thus, the characteristic quantity that can increase the weight of the pixel that approaches concerned pixel calculates, and particularly, under the less condition of noise ratio, can carry out noise level more accurately and detect.
And, in above-mentioned example, according to NR COEFFICIENT K f (x by NR handling part 36, y) the pixel value P (x to concerned pixel, and the LPF value Pf (x of two-dimensional directional y), y) be weighted addition, but, the invention is not restricted to this, also can replace the LPF value Pf (x, y) of two-dimensional directional, in weighting summation, use the one dimension LPF value (La (x of the direction of for example difference absolute value minimum, y) any one party in～Ld (x, y)).In this situation, compare with the situation of the LPF value Pf (x, y) that uses two-dimensional directional, become and use compared with the weighting summation of the filter value of close limit, therefore, the in the situation that in input picture, noise ratio being less, can maintain especially resolution completely, be effective in this respect.
In a word, only require out the LPF value (periphery low-pass filter values) of use concerned pixel and neighboring pixel thereof, for weighting summation NR handling part 36 and typical value concerned pixel.
, for processing window, output is used the low-pass filter values of concerned pixel and surrounding pixel thereof as periphery low-pass filter values to periphery low pass filter 25.
And, in above-mentioned example, mean value calculation in the formula (1) of the LPF value for obtaining two-dimensional directional is processed, use the same color pixel in the processing window being formed by 17 × 17 pixels to carry out computing, but, the invention is not restricted to this, also can use other window sizes than waiting according to the S/N of the installation cost that will realize and the picture signal that will input.And, in this situation, take advantage of the selection of such pixel count by carrying out power that the denominator of formula (1) is 2, can cut down general divider (divisor is the divider that any value all can be tackled).
And, in above-mentioned example, in processing, the NR of use formula (3) carries out the division arithmetic based on 64, and still, also can utilize the value beyond 64 to carry out division arithmetic.But, if the power that divisor is 2 is taken advantage of, utilizing in hard-wired situation, can be shifted and carry out division arithmetic by bit, do not need general divider, in the situation that utilizing software to realize, can improve the execution speed of computing.
And, the median that use is obtained by median filter 30 is as the typical value of concerned pixel, but, the pixel value of the concerned pixel the pixel value of the concerned pixel 1 frame that also can export according to the pixel value of the concerned pixel of the present frame the view data of exporting from line storage 10, from the 1st frame memory 11,2 frames of exporting from the 2nd frame memory 12 is obtained mean value, sets it as the typical value of concerned pixel.,, in this situation, typical value is mean value.Therefore, typical value calculating part calculating mean value.
And, also can use the filtering result as time-axis direction beyond median, mean value and the value that obtains as above-mentioned typical value.The value that for example, also can obtain use recursive filter on time-axis direction is as the typical value of concerned pixel.,, in this situation, typical value is the value being calculated by recursive filter.Therefore, typical value calculating part is recursive filter.
And then, obtain the pixel value before pixel value, 1 frame of present frame, the pixel value before 2 frames, but, also can also use the pixel value of the above past frame before of 3 frames to obtain typical value, in a word, as long as the value obtaining after the filtering of the time-axis direction of the pixel of the concerned pixel in 1 above frame beyond pixel value and the present frame of the concerned pixel of utilization use present frame and identical two-dimensional coordinate.
In addition, the motion (shake) that produces image in interframe, also can, after motion is compensated, carry out the filtering of time-axis direction.For example, the pixel (pixel corresponding with concerned pixel) of residing position (motion side) in the concerned pixel (representing the pixel of the image section identical with the concerned pixel of present frame) that uses the concerned pixel of present frame and be estimated as present frame in the past 1 or 2 frames above, obtains their typical value.In a word, as long as the concerned pixel in use present frame and the pixel corresponding with above-mentioned concerned pixel in past frame.
And, replace, producing motion in the situation that, also can not use the frame beyond present frame concerned pixel pixel value and use the pixel value of concerned pixel of present frame as typical value.
Execution mode 2
Fig. 9 illustrates the functional structure of the image processing apparatus of embodiments of the present invention 2.In Fig. 9, to the structural element mark same numeral identical with Fig. 1.
The structural element identical with execution mode 1 is line storage 10, frame memory 11,12, median filter 30, one-dimensional filtering device 21,22,23,24, two dimensional filter 25, noise level test section 32 and NR handling part 36.
Execution mode 2 is with the difference of execution mode 1, and correction coefficient generating unit 38 is set, and replaces the NR coefficient generating unit 34 of Fig. 1 and NR coefficient generating unit 44 is set.
NR coefficient generating unit 44 has the NR coefficient calculations 44a of portion, the NR coefficient correction 44b of portion.
The same formation of NR coefficient generating unit 34 of the NR coefficient calculations 44a of portion and Fig. 1, profit uses the same method and generates NR COEFFICIENT K f (x, y).
The NR coefficient correction 44b of portion is according to the correction coefficient hy (x exporting from correction coefficient generating unit 38, y) to the NR COEFFICIENT K f (x from the NR coefficient calculations 44a of portion output, y) proofread and correct, generate the NR COEFFICIENT K fb (x, y) after proofreading and correct.
Particularly, as shown in following formula (4), by NR COEFFICIENT K f (x, y) being multiplied by correction coefficient hy (x, y) and divided by 64, the NR COEFFICIENT K fb (x, y) after calculation correction.
Kfb(x，y)＝Kf(x，y)×hy(x，y)/64…(4)
, NR coefficient generating unit 44, according to the noise level NL (x, y) being detected by noise level test section 32 and the correction coefficient hy (x, y) being generated by correction coefficient generating unit 38, generates NR COEFFICIENT K fb (x, y).By according to the NR COEFFICIENT K fb (x, y) after formula (4) calculation correction, correction coefficient hy (x, y) is less, and the NR COEFFICIENT K fb (x, y) after proofreading and correct is less.
; correction coefficient hy (x; y) at periphery low-pass filter values Pf (x; y) lower than in the scope of setting along with periphery low-pass filter values Pf (x; reducing and reducing y), correction coefficient hy (x, y) is less; noise reduction coefficient Kf (x, y) is less.
Correction coefficient generating unit 38 is according to the output Pf (x by two-dimentional LPF25, y) lightness (lightness of each color component or intensity) of the image of neighboring area that represent, centered by each pixel, generate correction coefficient hy (x, y).
As the value of output Pf (x, y) and the relation (input-output characteristic) of correction coefficient hy (x, y) of two-dimentional LPF25 that is input to correction coefficient generating unit 38, for example, use the relation shown in Figure 10.In illustrated example, with respect to the two-dimentional LPF value Pf (x that gets 0～4095 value, y), if 129 stages that the value that correction coefficient is desirable is 0～128, do not proofreading and correct (COEFFICIENT K fb (the x after correction, y) with proofread and correct before COEFFICIENT K f (x, y) equate) situation under, the value of correction coefficient becomes 64.This be because, as mentioned above, by the value of correction coefficient is multiplied each other to proofread and correct divided by 64 values that obtain and NR COEFFICIENT K f (x, y).
In Figure 10, symbol NR1HX and NR1HY illustrate the predefined value for regulation input-output characteristic.Wherein, value NR1HX is the higher limit that determines the upper limit of the scope that NR coefficient is proofreaied and correct, and higher limit NR1HX is larger, wider for the correcting range of NR coefficient.Value NR1HY determines the value of degree of correction, and the value of making NR1HY becomes less value, stronger for making NR coefficient become the degree of correction of less value.
In the case of using the input-output characteristic of Figure 10, be less than higher limit NR1HX at two-dimentional LPF value Pf (x, y), in the dark portion of image, establishing the NR coefficient of exporting from NR coefficient generating unit 44 is less value, can weaken Noise Reduction.
In addition, also can replace as shown in figure 10 and make correction coefficient hy (x, y) become the value that is less than 64 in dark portion, and for example, make correction coefficient hy (x, y) become the value that is greater than 64 in bright portion (being greater than in the scope of regulation lower limit).And then, also can make correction coefficient hy (x, y) become less value in dark portion, and make correction coefficient hy (x, y) become the value that is greater than 64 in bright portion.
As mentioned above, according to this execution mode 2, in bright portion (part that each color is stronger) that can be in image and dark portion (part that each color is weak), change the degree of noise reduction, can prevent that the image quality that the difference of the lightness (color intensity) of input picture causes from reducing, for example, in the case of adjusting NR coefficient in conjunction with the bright areas of input picture, the part (dark portion) that signal level is lower thickens etc.
In addition, in the example of this execution mode 2, can switch based on two-dimentional LPF value Pf (x with 129 stages, y) correction, but, the invention is not restricted to this, the number of stages of switching for example can change according to the characteristic of the input picture that depends on imaging apparatus characteristic.And in formula (4), the value that correction coefficient hy (x, y) is obtained divided by 64 and COEFFICIENT K f (x, y) multiply each other, and still, the invention is not restricted to this, also can be divided by the value beyond 64.But, if the power that the value that makes to use in division arithmetic is 2 is taken advantage of, can be shifted and obtain (the x with COEFFICIENT K f by bit, y) value multiplying each other, in the situation that utilizing hardware to process, can cut down general divider, therefore can simplify circuit scale, and, in the situation that utilizing software to process, can realize the high speed of processing speed.
And, in the example of execution mode 2, for the image that uses single-plate color transducer to photograph, to whole shooting color, the correction of the pixel value (intensity of each color) based on each color is carried out in unification, but, the invention is not restricted to this, also can be configured to and utilize different input-output characteristics to generate correction coefficient according to each shooting color (each color component).Thus, according to the characteristic of the image input system of each color component or image delivery system, can carry out best noise reduction process according to the characteristic of input picture.
In order to generate correction coefficient according to each color component, same with the situation of explanation in execution mode 1, can distinguish LPF21～25, median filter 30, noise level test section 32, NR coefficient generating unit 34, NR handling part 36 and correction coefficient generating unit 38 that independent (according to each shooting color) arranges Fig. 9, also can utilize same circuits to process in order.
Execution mode 3
Figure 11 illustrates the functional structure of the image processing apparatus of embodiments of the present invention 3.In Figure 11, to the structural element mark same numeral identical with Fig. 9.
The structural element identical with execution mode 2 (Fig. 9) is line storage 10, frame memory 11,12, median filter 30, one-dimensional filtering device 21,22,23,24 and two dimensional filter 25.
The image processing apparatus of execution mode 3 replaces noise level test section 32, NR handling part 36 and the NR coefficient generating unit 44 of Fig. 9 and has noise level test section 37, NR handling part 39 and NR coefficient generating unit 45.Noise level test section 37, NR handling part 39 and NR coefficient generating unit 45 are roughly the same with noise level test section 32, NR handling part 36 and NR coefficient generating unit 44 respectively, still, have following difference.Especially, as described in detail below, NR handling part 39 is with the difference of NR handling part 36, with reference to the output of the 1st one dimension LPF～4th one dimension LPF.
And in execution mode 2, the NR coefficient calculations 44a of portion only calculates NR COEFFICIENT K f (x, y), still, in execution mode 3, the NR coefficient calculations 45a of portion calculates NR COEFFICIENT K g1 (x, y)～these 5 kinds of NR coefficients of Kg5 (x, y).NR COEFFICIENT K g1 (x, y) be the coefficient for the output valve of the 1st one dimension LPF, NR COEFFICIENT K g2 (x, y) be the coefficient for the output valve of the 2nd one dimension LPF, NR COEFFICIENT K g3 (x, y) is the coefficient for the output valve of the 3rd one dimension LPF, NR COEFFICIENT K g4 (x, y) be the coefficient for the output valve of the 4th one dimension LPF, NR COEFFICIENT K g5 (x, y) is the coefficient for the output valve of two-dimentional LPF.
In the noise level test section 37 of execution mode 3, according to formula (2a)～(2d) calculating difference absolute value ADa (x, y)～ADd (x, y), minimum value in the difference absolute value that output calculates is as the noise level NL1 (x of concerned pixel, y), output the 2nd little value is as noise level NL2 (x, y).
And, export the direction of the one dimension LPF corresponding with noise level NL1 (x, y) as DR1 (x, y).Any one party in DR1 (x, y) in the numbering (1～4) of storage one dimension LPF.
; in noise level test section 37; calculate each one-dimensional square to low-pass filter values and the absolute value ADa (x of the difference of typical value; y)～ADd (x; y); absolute value ADa (the x of the difference that output calculates; y)～ADd (x; y) in, meet the minimization of the sum of absolute value of rated condition, as the noise level NL1 (x, y) of concerned pixel; export the one-dimensional square corresponding with it to the direction of low pass filter as DR1 (x; y), output the 2nd little value is as the noise level NL2 (x, y) of concerned pixel.
In the NR of the execution mode 3 coefficient calculations 45a of portion, according to noise level NL1 (x, y), NL2 (x, y), direction DR1 (x, y), calculate the NR COEFFICIENT K g1 (x for the output valve of the 1st one dimension LPF, y), for the NR COEFFICIENT K g2 (x of the output valve of the 2nd one dimension LPF, y), for the NR COEFFICIENT K g3 (x of the output valve of the 3rd one dimension LPF, y), for the NR COEFFICIENT K g4 (x of the output valve of the 4th one dimension LPF, y) and for the NR COEFFICIENT K g5 (x, y) of the output valve of two-dimentional LPF.
In the NR coefficient calculations 45a of portion, according to minimum value NL1 (x, y) and the one-dimensional square corresponding with it to the direction DR1 (x of low pass filter, y), described the 2nd little value NL2 (x, y), generate for pixel value, periphery low-pass filter values Pf (x, y), multiple one dimension low-pass filter values La (x to concerned pixel, y)～Ld (x, y) is weighted the noise reduction coefficient of addition.
In the calculating of NR COEFFICIENT K g1 (x, y)～Kg5 (x, y), NR coefficient corresponding the one dimension LPF initial different with direction DR1 (x, y) with direction is all set as to 0.For example, be 1 at direction DR1 (x, y), NR COEFFICIENT K g2 (x, y)～Kg4 (x, y) is set as 0.
Then, calculate for to the output valve of one dimension LPF corresponding to direction DR1 (x, y) and the output valve of two-dimentional LPF between carry out the 1st interpolation coefficient KL1 (x, y) of interpolation.As shown in figure 12, with reference to the noise level NL2 (x, y) and the ratio (ratio of the former with the latter) of noise level NL1 (x, y) of direction that becomes minimum value of direction that becomes the 2nd little value, calculate the 1st interpolation coefficient KL1 (x, y).Be below R1min in the situation that at described ratio, for example, if the 1st interpolation coefficient KL1 (x, y) is minimum value (0), be that R1max is above in the situation that at described ratio, for example, if the 1st interpolation coefficient KL1 (x, y) is maximum (64).Then, between R1min and R1max in the situation that, establish the 1st interpolation coefficient KL1 (x, y) for the linear value increasing between minimum value and maximum at described ratio.
Then, calculate the 2nd interpolation coefficient KL2 (x, y) for carrying out interpolation between the output valve to one dimension LPF or two-dimentional LPF and the pixel value of concerned pixel.As shown in figure 13, becoming the noise level NL1 (x of direction of minimum value, y) be in the situation below R2MINX, if the 2nd interpolation coefficient KL2 (x, y) be R2MINY, at the noise level NL1 (x, y) of direction that becomes minimum value for more than R2MAXX, if the 2nd interpolation coefficient KL2 (x, y) is R2MAXY.Between R2MINX and R2MAXX in the situation that, be made as the linear value increasing between R2MINY and R2MAXY at the noise level NL1 (x, y) of direction that becomes minimum value.
Finally, calculate the NR COEFFICIENT K gN (x corresponding with the one dimension LPF of direction that becomes minimum value according to formula (5), y) (N is any one party in 1～4), calculates the NR COEFFICIENT K g5 (x, y) corresponding with two-dimentional LPF according to formula (6).
KgN(x，y)
＝KL1(x，y)×KL2(x，y)÷64…(5)
Kg5(x，y)
＝(64-KL1(x，y))×KL2(x，y)÷64…(6)
In the NR of the execution mode 3 coefficient correction 45b of portion, for NR COEFFICIENT K g1 (x, y)～Kg5 (x, y), be multiplied by correction coefficient hy (x, y) after divided by 64 same with formula (4), respectively the NR COEFFICIENT K gb1 (x after calculation correction respectively, y)～Kgb5 (x, y).
In the NR of execution mode 3 handling part 39, carry out the weighting summation of the output valve of the pixel value of concerned pixel, the output valve of whole one dimension LPF, two-dimentional LPF according to formula (7).
Pb(x，y)
＝Kgb1(x，y)×La(x，y)+Kgb2(x，y)×Lb(x，y)
+Kgb3(x，y)×Lc(x，y)+Kgb4(x，y)×Ld(x，y)
+Kgb5(x，y)×Pf(x，y)
+(1-Kgb1(x，y)-Kgb2(x，y)-Kgb3(x，y)
-Kgb4(x，y)-Kgb5(x，y))×P(x，y)
…(7)
Carry out pixel value Pb (x, y) after as noise reduction process of the value Pb (x, y) that obtains after the computing of formula (7) and export from NR handling part 39.
,, in NR handling part 39, according to noise reduction coefficient, the pixel value to concerned pixel, periphery low-pass filter values, multiple one dimension low-pass filter values are weighted addition.
As image processing apparatus, the present invention has been described above, still, the image processing method of being implemented by this image processing apparatus also forms a part of the present invention.And then, also form a part of the present invention for the image processing program that makes computer carry out this image processing method.
In addition, embodiments of the present invention are described as mentioned above, still, have the invention is not restricted to these execution modes.
Label declaration
10: line storage; 11: the 1 frame memories; 12: the 2 frame memories; 21～24: one dimension LPF; 25: two-dimentional LPF; 32,37: noise level test section; 34:NR coefficient generating unit; 36,39:NR handling part; 38: correction coefficient generating unit; 30: median filter; 44,45:NR coefficient generating unit; 44a, 45a:NR coefficient calculations portion; 44b, 45b:NR coefficient correction portion.
Claims (14)
1. an image processing apparatus, is characterized in that,
Described image processing apparatus is using the dynamic image being made up of the time series of multiframe as input picture,
Described image processing apparatus has:
Multiple one dimension low pass filters, they calculate the low-pass filter values of the pixel that uses described concerned pixel and upwards arrange at mutual different one-dimensional square respectively with respect to described concerned pixel in the processing window being made up of multiple pixels centered by the concerned pixel of paying close attention to frame;
Periphery low pass filter, it uses the low-pass filter values of described concerned pixel and surrounding pixel thereof for described processing window output, as periphery low-pass filter values;
Typical value calculating part, it is according to the pixel value of the pixel of position corresponding with described concerned pixel in 1 beyond the pixel value of the described concerned pixel of described concern frame and described concern frame above frame, carry out the filtering of time-axis direction, obtain thus the typical value relevant with described concerned pixel;
Noise level test section, its calculate one-dimensional square described in each to low-pass filter values and the absolute value of the difference of described typical value, in the absolute value of the described difference that output calculates, meet the minimization of the sum of absolute value of rated condition, as the noise level of concerned pixel;
NR coefficient generating unit, it receives described noise level, generates the noise reduction coefficient corresponding with this noise level according to predefined input-output characteristic; And
NR handling part, it uses pixel value and the described periphery low-pass filter values of described noise reduction coefficient to described concerned pixel to be weighted addition,
In described NR handling part, described noise reduction coefficient is larger, larger for the weight of described periphery low-pass filter values.
2. image processing apparatus according to claim 1, is characterized in that,
Described typical value is median.
3. image processing apparatus according to claim 1, is characterized in that,
Described typical value is mean value.
4. image processing apparatus according to claim 1, is characterized in that,
Described typical value is the value being calculated by recursive filter.
5. according to the image processing apparatus described in any one in claim 1～4, it is characterized in that,
Described periphery low-pass filter values is two-dimensional low-pass filter value.
6. according to the image processing apparatus described in any one in claim 1～5, it is characterized in that,
In described processing window, only select with the difference of described typical value be positioned at regulation level difference pixel value and obtain each low-pass filter values, thereby obtain described one dimension low pass filter and described periphery low pass filter.
7. according to the image processing apparatus described in any one in claim 1～6, it is characterized in that,
Described image processing apparatus also has correction coefficient generating unit, and this correction coefficient generating unit generates correction coefficient according to described periphery low-pass filter values,
The correction coefficient that described NR coefficient generating unit generates according to the noise level being detected by described noise level test section with by described correction coefficient generating unit, generates described NR coefficient.
8. image processing apparatus according to claim 7, is characterized in that,
Described correction coefficient, reduces along with reducing of described periphery low-pass filter values in the scope lower than setting in described periphery low-pass filter values,
Described correction coefficient is less, and described noise reduction coefficient is less.
9. according to the image processing apparatus described in any one in claim 1～8, it is characterized in that,
Described noise level test section calculate one-dimensional square described in each to low-pass filter values and the absolute value of the difference of described typical value, in the absolute value of the described difference that calculates of output, meet the minimization of the sum of absolute value of rated condition and the one-dimensional square corresponding with it direction and the 2nd little value to low pass filter, as the noise level of concerned pixel
Described NR coefficient generating unit according to described minimum value and the one-dimensional square corresponding with it to the direction of low pass filter and described the 2nd little value, generate the noise reduction coefficient that is weighted addition for pixel value, described periphery low-pass filter values, described multiple one dimension low-pass filter values to described concerned pixel
Described NR handling part is according to described noise reduction coefficient, and the pixel value to described concerned pixel, described periphery low-pass filter values, described multiple one dimension low-pass filter values are weighted addition.
10. according to the image processing apparatus described in any one in claim 1～9, it is characterized in that,
Described input picture comprises multiple color components,
Described NR coefficient calculations portion, according to each color component of described input picture, utilizes the input-output characteristic corresponding with the characteristic of described input picture to generate the noise reduction coefficient of each color component.
11. according to the image processing apparatus described in any one in claim 7～9, it is characterized in that,
Described input picture comprises multiple color components,
Described correction coefficient generates according to each color component of described input picture,
Utilize according to the characteristic of described input picture and different input-output characteristics generates described correction coefficient.
12. 1 kinds of image processing methods, is characterized in that,
Using the dynamic image being formed by the time series of multiframe as input picture,
Described image processing method has following steps:
Multiple one dimension low-pass filtering steps, in the processing window being formed by multiple pixels centered by the concerned pixel of paying close attention to frame, calculate the low-pass filter values of the pixel that uses described concerned pixel and upwards arrange at mutual different one-dimensional square respectively with respect to described concerned pixel;
Periphery low-pass filtering step, for the low-pass filter values of the described processing window output described concerned pixel of use and surrounding pixel thereof, as periphery low-pass filter values;
Typical value calculation procedure, according to the pixel value of the pixel of position corresponding with described concerned pixel in 1 beyond the pixel value of the described concerned pixel of described concern frame and described concern frame above frame, obtains the typical value relevant with described concerned pixel;
Noise level detecting step, calculate one-dimensional square described in each to low-pass filter values and the absolute value of the difference of described typical value, in the absolute value of the described difference that output calculates, meet the minimization of the sum of absolute value of rated condition, as the noise level of concerned pixel;
NR coefficient generates step, receives described noise level, generates the noise reduction coefficient corresponding with this noise level according to predefined input-output characteristic; And
NR treatment step, uses pixel value and the described periphery low-pass filter values of described noise reduction coefficient to described concerned pixel to be weighted addition,
In described NR treatment step, described noise reduction coefficient is larger, larger for the weight of described periphery low-pass filter values.
13. image processing methods according to claim 12, is characterized in that,
Described image processing method also has correction coefficient and generates step, and generate in step and generate correction coefficient according to described periphery low-pass filter values in this correction coefficient,
Generate in step at described NR coefficient, the correction coefficient generating according to the noise level detecting in described noise level detecting step with in described correction coefficient generation step generates described NR coefficient.
14. 1 kinds of image processing programs, wherein, this image processing program is for making computer execute claims the processing of each step of 12 or 13 image processing method.
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JPJP2012-085416 | 2012-04-04 | ||
| JP2012-085416 | 2012-04-04 | ||
| JP2012085416 | 2012-04-04 | ||
| PCT/JP2013/054318 WO2013150824A1 (en) | 2012-04-04 | 2013-02-21 | Image processing device and method, and image processing program |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104205805A true CN104205805A (en) | 2014-12-10 |
| CN104205805B CN104205805B (en) | 2017-07-04 |
Family
ID=49300324
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201380017115.5A Expired - Fee Related CN104205805B (en) | 2012-04-04 | 2013-02-21 | Image processing apparatus and method |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US9258461B2 (en) |
| EP (1) | EP2835963B1 (en) |
| JP (1) | JP5933690B2 (en) |
| CN (1) | CN104205805B (en) |
| WO (1) | WO2013150824A1 (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR3090258A1 (en) | 2018-12-12 | 2020-06-19 | Stmicroelectronics Sa | Method and device for estimating the noise level of reference darkness rows of an image sensor |
| KR20220148423A (en) * | 2021-04-29 | 2022-11-07 | 삼성전자주식회사 | Denoising method and denosing device of reducing noise of image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006010276A1 (en) * | 2004-07-30 | 2006-02-02 | Algolith Inc | Apparatus and method for adaptive 3d artifact reducing for encoded image signal |
| US20060245506A1 (en) * | 2005-05-02 | 2006-11-02 | Samsung Electronics Co., Ltd. | Method and apparatus for reducing mosquito noise in decoded video sequence |
| US20100060749A1 (en) * | 2008-09-09 | 2010-03-11 | Sujith Srinivasan | Reducing digital image noise |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE69329670T2 (en) * | 1992-09-14 | 2001-03-15 | Thomson Multimedia Sa | Method and device for noise reduction |
| JPH07134772A (en) | 1993-11-10 | 1995-05-23 | Ge Yokogawa Medical Syst Ltd | Image display device and image processing method |
| JP4069581B2 (en) | 2000-12-11 | 2008-04-02 | セイコーエプソン株式会社 | Image processing method, computer-readable recording medium on which image processing program is recorded, and image processing apparatus |
| JP3729118B2 (en) | 2001-10-23 | 2005-12-21 | ノーリツ鋼機株式会社 | Image processing method, image processing apparatus, image processing program, and computer-readable recording medium recording the same |
| JP3934506B2 (en) | 2002-08-06 | 2007-06-20 | オリンパス株式会社 | Imaging system and image processing program |
| US7437013B2 (en) * | 2003-12-23 | 2008-10-14 | General Instrument Corporation | Directional spatial video noise reduction |
| JP2007312304A (en) | 2006-05-22 | 2007-11-29 | Fujitsu Ltd | Image processing apparatus and image processing method |
| JP5253835B2 (en) | 2008-02-19 | 2013-07-31 | 株式会社キーエンス | Image generating apparatus, image generating method, and computer program |
| JP4990240B2 (en) * | 2008-08-06 | 2012-08-01 | 三菱電機株式会社 | Image processing apparatus and image processing program |
| JP5315158B2 (en) | 2008-09-12 | 2013-10-16 | キヤノン株式会社 | Image processing apparatus and image processing method |
| EP2348482B1 (en) | 2008-11-21 | 2015-06-17 | Mitsubishi Electric Corporation | Image processing device and image processing method |
| JP2011124694A (en) | 2009-12-09 | 2011-06-23 | Sony Corp | Image processing device, image processing system, and image pickup device |
| US8654259B2 (en) * | 2010-02-18 | 2014-02-18 | Sharp Kabushiki Kaisha | Image processing system and display device |
-
2013
- 2013-02-21 US US14/390,626 patent/US9258461B2/en not_active Expired - Fee Related
- 2013-02-21 CN CN201380017115.5A patent/CN104205805B/en not_active Expired - Fee Related
- 2013-02-21 EP EP13772701.2A patent/EP2835963B1/en not_active Not-in-force
- 2013-02-21 WO PCT/JP2013/054318 patent/WO2013150824A1/en active Application Filing
- 2013-02-21 JP JP2014509072A patent/JP5933690B2/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006010276A1 (en) * | 2004-07-30 | 2006-02-02 | Algolith Inc | Apparatus and method for adaptive 3d artifact reducing for encoded image signal |
| US20060245506A1 (en) * | 2005-05-02 | 2006-11-02 | Samsung Electronics Co., Ltd. | Method and apparatus for reducing mosquito noise in decoded video sequence |
| US20100060749A1 (en) * | 2008-09-09 | 2010-03-11 | Sujith Srinivasan | Reducing digital image noise |
Non-Patent Citations (1)
| Title |
|---|
| AMER A ET AL: "fast and reliable structure-oriented video noise estimation，XP011124672", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY,IEEE SERVICE CENTER,PISCATAWAY,NJ,US》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| JP5933690B2 (en) | 2016-06-15 |
| US20150054990A1 (en) | 2015-02-26 |
| US9258461B2 (en) | 2016-02-09 |
| EP2835963B1 (en) | 2017-03-22 |
| EP2835963A1 (en) | 2015-02-11 |
| WO2013150824A1 (en) | 2013-10-10 |
| JPWO2013150824A1 (en) | 2015-12-17 |
| CN104205805B (en) | 2017-07-04 |
| EP2835963A4 (en) | 2016-01-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP4054184B2 (en) | Defective pixel correction device | |
| US7283164B2 (en) | Method for detecting and correcting defective pixels in a digital image sensor | |
| JP5918915B2 (en) | Method and device for creating an improved color image using a sensor having a color filter | |
| US20150092079A1 (en) | Imaging systems and methods for generating motion-compensated high-dynamic-range images | |
| US8004586B2 (en) | Method and apparatus for reducing noise of image sensor | |
| US7884864B2 (en) | Noise reduction device, noise reduction method and image capturing device | |
| US20120163734A1 (en) | Apparatus for color interpolation using adjustable threshold | |
| US9959633B2 (en) | Texture detection in image processing | |
| US9031350B2 (en) | Method for processing edges in an image and image processing apparatus | |
| US7751642B1 (en) | Methods and devices for image processing, image capturing and image downscaling | |
| CN101473347A (en) | Picture enhancing increasing precision smooth profiles | |
| US7239758B2 (en) | Signal processing device for reducing noise of image signal, signal processing program, and signal processing method | |
| CN101534375B (en) | Method for correcting chromatic aberration | |
| CN110418081B (en) | High dynamic range image full-resolution reconstruction method and device and electronic equipment | |
| US6876778B2 (en) | Edge enhancement method and apparatus in digital image scalar-up circuit | |
| CN104205805A (en) | Image processing device and method, and image processing program | |
| US20090010561A1 (en) | Device for removing noise in image data | |
| US20140355872A1 (en) | Method for determining interpolating direction for color demosaicking | |
| US20090207268A1 (en) | Image signal processing device, image signal processing method and imaging device | |
| US8054348B2 (en) | Noise reduction device and digital camera | |
| US20190387205A1 (en) | Tile-selection based deep demosaicing acceleration | |
| JP3149729B2 (en) | Block distortion removal device | |
| KR101327790B1 (en) | Image interpolation method and apparatus | |
| JP2008028475A (en) | Image processor, image processing program, recording medium recording image processing program, image processing method | |
| CN109816597B (en) | Method and device for removing GrGb noise of image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170704 Termination date: 20200221 |